High-Rate Phase Association with Travel Time Neural Fields

核心概念

Harpa introduces a novel framework for high-rate seismic phase association, overcoming challenges posed by unknown wave speeds and dense seismic events.

要約

• The abstract highlights the importance of associating arrival phases with originating earthquakes.
• Existing techniques struggle in high-rate regimes with unknown wave speeds.
• Harpa framework utilizes deep generative modeling and neural fields to address these challenges.
• Results show Harpa's accuracy in seismic phase association in complex scenarios.
• The method is robust to distorted wave speeds and out-of-distribution scenarios.
• Harpa demonstrates resilience to spurious and missing picks in seismic data.
• The study opens new avenues for understanding seismicity in dense and high-rate seismic events.

統計

Deep-learning-based phase detection detects small, high-rate arrivals from seismicity clouds.
Harpa is the first seismic phase association framework accurate in the high-rate regime.
The method is robust to unknown wave speeds and distorted wave speed models.

引用

"Our understanding of regional seismicity relies on associating arrival phases with originating earthquakes."
"Harpa paves the way for new avenues in exploratory Earth science and improved seismicity understanding."

抽出されたキーインサイト

High-Rate Phase Association with Travel Time Neural Fields

by Chen... 場所 arxiv.org 03-28-2024

https://arxiv.org/pdf/2307.07572.pdf

深掘り質問

How can Harpa's framework be applied to real-time seismic monitoring systems

Harpa's framework can be applied to real-time seismic monitoring systems by leveraging its ability to handle dense seismic event data with unknown wave speeds. In a real-time monitoring system, Harpa can continuously process incoming seismic data from multiple stations, associating arrival phases with their originating earthquakes. By utilizing deep generative modeling and neural fields, Harpa can estimate earthquake locations, occurrence times, and wave speed models simultaneously. This real-time processing capability allows for quick and accurate identification and association of seismic events, providing valuable insights into earthquake dynamics as events unfold. Additionally, the use of stochastic optimization methods like Stochastic Gradient Langevin Dynamics (SGLD) enables efficient exploration of the loss landscape, facilitating timely and accurate results in a real-time monitoring setting.

What are the limitations of Harpa in scenarios with extremely high rates of seismic events

One limitation of Harpa in scenarios with extremely high rates of seismic events is the computational complexity associated with processing a large number of simultaneous arrivals. In such scenarios, the dense occurrence of seismic events can lead to overlapping arrivals at different stations, making the association task more challenging. The complexity of the optimization problem increases with the number of events and stations, potentially leading to longer processing times and higher computational resource requirements. Additionally, the presence of spurious or missing picks in high-rate scenarios can introduce noise and errors in the association process, affecting the overall accuracy of the results. Therefore, while Harpa excels in handling dense seismic data, its performance may be impacted in scenarios with extremely high rates of seismic events due to the increased computational and processing demands.

How can the principles of Harpa be adapted to other fields beyond seismology

The principles of Harpa can be adapted to other fields beyond seismology by applying the concept of deep generative modeling and neural fields to similar inverse problems in different domains. For example, in medical imaging, Harpa's framework could be utilized for reconstructing images from noisy or incomplete data, such as in MRI or CT scans. By treating the observed data as probability measures and leveraging optimal transport techniques, Harpa's approach can be extended to tasks like image reconstruction, object tracking, or anomaly detection in various fields. The use of deep learning and stochastic optimization methods can enhance the robustness and efficiency of solving inverse problems in diverse applications, making Harpa's principles adaptable to a wide range of domains requiring data reconstruction and association tasks.